Cost of Opioid-Related Adverse Drug Events in Surgical Patients

276

Journal of Pain and Symptom Management

Vol. 25 No. 3 March 2003

Original Article

Cost of Opioid-Related Adverse Drug Events in Surgical Patients Gary M. Oderda, PharmD, MPH, R. Scott Evans, PhD, James Lloyd, BS, Arthur Lipman, PharmD, Connie Chen, PharmD, Michael Ashburn, MD, MPH, John Burke, MD, and Matthew Samore, MD University of Utah College of Pharmacy and School of Medicine (G.O., R.S.E., A.L., M.A., J.B., M.S.), and LDS Hospital (R.S.E., J.L., J.B.), Salt Lake City, Utah; Pharmacia Corp. (C.C.), Peapak, New Jersey; and Intermountain Health Care (R.S.E.), Salt Lake City, Utah, USA

Abstract Opioids have demonstrated efficacy and often are drugs of choice in the management of postoperative pain. However, their use is often limited by adverse drug events (ADEs). The objective of this study was to determine the ADE rate in adult surgical patients who received opioids and the impact of opioid ADEs on length of stay (LOS), costs, and mortality. A hospitalbased computerized system detected potential ADEs. Adult patients were selected if they received at least one dose of opioid medication during a surgical hospitalization between 1 January 1990 and 31 December 1999. Control patients were matched based on matching length of stay ([LOS] at least as long as time to ADE), age (within 10 years), sex, admission year, major disease category (MDC), and without an ADE. Linear regression models were used to determine the predictors of increased LOS, total hospital costs, and log-transformed total hospital costs. 60,722 patients received opioid medication during their surgical hospitalization and 2.7% experienced an opioid-related ADE. The most common clinical manifestations were nausea and vomiting (67%), and rash, hives, or itching (33.5%). No statistically significant difference was seen in mortality between ADE/non-ADE patients. ADE patients had statistically significant increases in LOS (0.53 days) and in log-transformed cost (16%). The estimated log cost difference of 16%, if applied to the median cost patient in the non-ADE group, averaged US$ 840. Opioidrelated ADEs are common in hospitalized patients and increase LOS and total hospital costs. J Pain Symptom Manage 2003;25:276–283. © 2003 U.S. Cancer Pain Relief Committee. Published by Elsevier. All rights reserved. Key Words Pain, opioids, adverse events, pharmacoeconomics

Introduction Opioids have demonstrated efficacy for pain relief after surgery and often are the analgesics

Address reprint requests to: Gary M. Oderda, PharmD, MPH, Department of Pharmacy Practice, University of Utah, 30 S 2000 E, Room 258, Salt Lake City, UT 84112, USA. Accepted for publication: May 17, 2002. © 2003 U.S. Cancer Pain Relief Committee Published by Elsevier. All rights reserved.

of choice for perioperative pain, but their use to achieve maximal pain relief is often accompanied by adverse drug events (ADEs). Some adverse events reflect medical errors. As a recent Institute of Medicine Report points out, medical errors are an important cause of morbidity, mortality, and cost in the United States Health Care System.1 Other adverse events do not reflect error, but are inherent in the pharmacology of the drugs. For all drugs, the inci0885-3924/03/$–see front matter doi:10.1016/S0885-3924(02)00691-7

Vol. 25 No. 3 March 2003

Cost of Opioid Adverse Drug Events

dence of serious and fatal ADEs in U.S. hospitals is as high as 6.7% and 0.32%, respectively.2 In 1992, the U.S. Department of Health and Human Services published a detailed, evidencebased clinical practice guideline for the management of acute, postoperative pain. This guideline advocates the use of postoperative opioid analgesia and documents adverse outcomes associated with inadequate postoperative pain management.3 The American Pain Society Principles of Analgesic Use also advocates the use of opioids in the management of perioperative pain.4 Opioids are associated with a relatively high incidence of ADEs and require careful monitoring and dose titration. Moreover, even careful dose titration cannot prevent some of the toxicities associated with opioid analgesia. Thus, it is important to consider the patients’ recovery experience and cost of adverse events resulting from postoperative opioid analgesia in order to develop guidelines for optimal, cost-effective patient care. As a drug class, analgesics contribute substantially to the overall burden of ADEs in hospitalized patients. For instance, Bates et al. reported that 29% of preventable ADEs were associated with analgesics, the largest share of any medication type, and in a previous study from LDS Hospital, morphine was identified as the single most common drug associated with an adverse event.5,6 Globally, costs associated with ADEs are very large, estimated to be in the range of 17 to 29 billion dollars per year in the U.S.7 Bates et al. found that there was a statistically significant difference in length of stay (2.2 days), charges (US$ 6,341) and costs (US$ 3,244) in patients with an ADE. Differences were even greater for preventable ADEs. The objectives of this study were a) to determine the incidence of opioid-related ADEs in adult surgery patients admitted to LDS Hospital from 1 January 1990 to 31 December 1999 and b) to quantify the effect of opioid-related ADEs on LOS, hospital costs, and mortality in this population.

Methods Study Population and ADE Surveillance Method LDS Hospital is a 520-bed teaching hospital affiliated with the University of Utah School of

277

Medicine. In 1989, the HELP (Health Evaluation through Logical Processing)8 hospital information system was used to develop a series of computer programs to improve the detection of ADEs at LDS Hospital. The core of the surveillance method was the use of computer signals to flag patients who had a possible ADE. Laboratory test results, drug levels, antidote drug orders, nurse charting, and other data in the electronic medical record were processed using computer logic. Reports were generated daily on patients who triggered the computer signals. A designated pharmacist reviewed those reports and collected additional data, as needed, from clinical personnel or the patient, to determine whether an ADE had occurred. The pharmacist applied the Naranjo criteria to each possible event, in order to assess the likelihood that the clinical manifestations were drugrelated.9 The Naranjo method uses a standardized set of questions to classify the ADE in relation to the suspect drug or relationship to a drug as either definite, probable, possible, or negative. Information on each ADE and computer signal were stored in a relational database; data elements included the causative drug, ADE type, severity, symptoms, service, ADE date, and admission date.

Selection of the Matched Data Set A matched cohort study design was followed. The study base population consisted of all inpatients aged greater than 17 years of age who were admitted to LDS Hospital from 1 January 1990 through 31 December 1999, received at least one dose of an opioid, and underwent a surgical procedure. The opioid was defined as a drug product that contained codeine, fentanyl, meperidine, methadone, morphine, propoxyphene, or oxycodone. The database was linked to the ADE database using the patient encounter number to identify surgical patients who experienced a verified ADE. Data extracted from the electronic medical record were also managed in a relational database. Each ADE patient was matched with up to five non-ADE patients who were randomly selected from LDS Hospital patients and met the following criteria: age within ten years, LOS at least as long as the time from admission to ADE for cases, same sex, major disease category (MDC),10 and year of admission.

278

Oderda et al.

A cost-based microcosting system was used to determine true hospital costs.11 This system uses time and motion studies to derive cost estimates for each type of service. In this study, cost is defined as total hospital costs. The University of Utah and LDS Hospital Institutional Review Boards approved the study.

Statistical Analysis The cumulative incidence of opioid-related ADEs was calculated for the entire cohort and for subgroups defined on the basis of type of surgical procedure. Linear regression was used to estimate the effect of ADE on cost and LOS. The independent variable of interest was the occurrence of an ADE. A dummy variable for each matched group was incorporated into regression models to account for matching. Fixed effect linear regression models that include dummy variables for each matched group, and a variable representing the occurrence of ADE, were used and are comparable to a paired t-test of ADE versus non-ADE patients. Random effect models also were constructed, which carry the alternative assumption that the matched groups were not fixed but rather derive from a population.12 Although the random effects model is generally considered more conservative statistically, we believe that both are acceptable approaches to apply to these data. To control for confounding by variables not included in the matching process, additional variables were added to the regression models on the basis of significant confounding (15% change in coefficient for ADE) or on statistical significance (P  0.05) by the Student t-test. The additional variables tested included DRG, DRG weight, and surgery type. Interactions between ADE occurrence and other independent variables also were examined. Additional subgroup analyses were performed according to type of surgery and ADE clinical manifestation. Because the outcome measures of LOS and hospital costs were highly skewed and non-normal, analyses of log transformed costs and LOS were also performed. In the log-transformed models the natural log of cost or LOS was calculated. Stata 7.0 and SAS 8.0 were used for analyses and statistical evaluations.13,14

Vol. 25 No. 3 March 2003

Results Descriptive Analysis A total of 4,452 ADEs were detected during the ten-year study period, of which 59% involved an opioid. The study cohort consisted of 60,722 hospitalized surgical patients who received at least one dose of an opioid analgesic. By prospective ADE surveillance, 1,623 (3%) of these patients experienced an opioid-related ADE; 1,612 (99%) of 1,623 ADEs were moderate in severity. On the basis of the Naranjo criteria, 79% were considered probable and 21% definite. Overall, 95% of opioid ADEs were due to injectable opioids. The opioid drugs most often associated with ADE were morphine (1,173 cases, 74%) and meperidine (303 cases, 19%). This most likely reflects higher use of morphine vs. meperidine at this hospital and does not suggest an increased incidence of morphine ADEs. Table 1 compares characteristics of surgical patients with and without an opioid ADE. The risk of an opioid-associated ADE varied based on type of surgery. ADEs secondary to opioids were most common in urology patients and least common in vascular surgery patients. No ADEs were identified in ophthalmology patients. Table 2 shows the risk of an opioid ADE by surgical service and compares the opioid ADEs to ADEs from other drugs in the entire surgical population. Across all surgery categories, the overall risk of having an opioid ADE was 2.6 times higher than the risk of having an ADE from all other drugs. The risk of an opioid ADE was higher in all surgery categories except the three with the lowest number of patients, ear nose and throat, vascular, and ophthalmology. Due to the small subset sample size in these three groups, there was insufficient power to compare ADE incidence across these subsets. A total of 1,586 (98%) of the ADE patients matched to at least one non-ADE patient and 92% matched to the maximum of 5 non-ADE patients. Thus, the final population for the matched analysis consisted of 1,586 cases and 7,696 matched controls. Sex, age, ADE severity, race, mortality, and LOS were compared in 37 ADE patients who did not match to those who did match. Statistically significant differences were seen only in age and LOS. Unmatched patients were younger, (46 vs 52 years [P  0.043]) and had a longer LOS, (14 vs 5.8 days [P  0.001]).

Vol. 25 No. 3 March 2003

Cost of Opioid Adverse Drug Events

279

Table 1 Characteristics of Surgical Patients with Opioid-Related Adverse Drug Events and Matched Surgical Patients Without Opioid-Related Adverse Drug Events Patients with ADE (n  1,586) Age, years Mean (standard deviation) Median (interquartile range) Women, n (%) Interval between admission and Adverse Drug Event, days Mean (standard deviation) Median (interquartile range) Opioid drug, n (%) Morphine Meperidine Oxycodone or Hydrocodone Fentanyl Other Most common major disease classifications, n (%) Female reproductive system Musculoskeletal system Digestive system Circulatory system Endocrine Most common diagnosis-related groups Major joint & limb reattachment procedures (209) Uterine & adnexa procedure for non-malignancy without CC (359) Major small & large bowel procedures (148) Uterine & adnexa procedure for non-malignancy with CC (358) OR procedures for obesity (288) LOS, days Mean (standard deviation) Median (interquartile range) Hospital costs ($) Mean (standard deviation) Median (interquartile range) Mortality, n (%)

Matched Analysis: Mortality and LOS A mortality rate of 0.7% (11/1,586) was seen in ADE patients as compared with 1% (78/ 7,696) in non-ADE patients. This difference was not statistically significant (conditional logistic regression, P  0.147). In the matched regression model for LOS that did not include other factors, an ADE was associated with a 0.66 day increase in hospital stay (P  0.001). After adjusting for type of surgery and DRG weight, an ADE was associated with a 0.53 day increase in LOS (P  0.01).

Matched Analysis: Cost Overall, mean costs per patient were $10,559 in the 1,586 ADE patients and $9,287 in the 7,696 matched non-ADE patients. A matched regression model was developed for both cost and log costs because of significant skew in the

52 (17) 50 (38–67) 1,076 (68) 1.8 (2.9) 1 (0–2)

Matched Non-ADE Patients (n  7,696) 52 (17) 51 (39–67) 5,247 (68) N/A N/A

1,173 (74) 303 (19) 67 (4) 28 (2) 2 (0)

2,845 (37) 1,825 (24) 2,721 (35) 59 (1) 246 (3)

372 (24) 364 (24) 241 (15) 110 (7) 106 (7)

1,855 (23) 1,813 (23) 1,193 (16) 540 (7) 512 (7)

151 (12) 148 (11) 98 (7) 98 (7) 93 (7)

556 (9) 879 (15) 365 (6) 408 (7) 392 (6)

6.7 (6.8) 4 (5–7) 10558 (14064) 6558 (4,429–11,436) 11 (0.7)

5.8 (6.8) 3 (4–7) 9287 (14065) 5253 (3,354–10,197) 78 (1)

distribution of costs (test for normality, P  0.00001). In the untransformed model, which did not include other factors in addition to matching, occurrence of an ADE was associated with a $714 increase in costs (P  0.026). The corresponding effect, when surgery type and DRG weight were added to the model, was $355 (P  0.225). In the log-transformed model, adjusted for surgery type and DRG weight, ADE was associated with a 16% increase in costs (P  0.001). Table 3 shows a summary of regression results. An analysis of outliers in the untransformed model revealed 6 patients who had total hospital costs in excess of $200,000. One of these patients was in the ADE group; five were not. If these patients were excluded from the nontransformed cost model that included surgery type and DRG weight, ADE was associated with $499 higher costs (P  0.057).

280

Oderda et al.

Vol. 25 No. 3 March 2003

Table 2 Risk of an Opioid-Versus Non-Opioid-Related ADE by Surgical Service

Surgery Category

Total Surgery Patients

Total Opioid Patientsa

Opioid-Related ADE Cases n (%)

Non-Opioid-Related ADE Cases n (%)

5,243 9,478 22,081 1,590 1,579 3,397 21,719 2,583 16,438 1,194 2,085 1,077

3,501 7,761 14,855 1,202 3,501 1,982 14,092 1,548 12,092 398 954 171

169 (4.8) 351 (4.5) 429 (2.9) 32 (2.7) 169 (2.6) 20 (2.5) 349 (2.5) 26 (1.7) 148 (1.2) 4 (1.0) 7 (0.7) 0 (0)

39 (0.7) 27 (0.3) 278 (1.3) 29 (1.8) 6 (0.4) 38 (1.1) 131 (0.6) 18 (0.7) 257 (1.6) 58 (4.9) 57 (2.7) 4 (0.4)

Urology OB-GYN General Transplant Other Plastic Orthopedics Neurosurgery Thoracic Ear nose and throat Vascular Ophthalmology aPatients

who had surgery and received at least one dose of an opioid.

The estimated log cost difference of 16%, if applied to the median cost patient in the nonADE group, amounts to $840. Using the $840 in costs attributable to opioid ADEs, the 1,623 opioid ADEs identified during the study period resulted in $1,363,322 in additional cost to LDS Hospital during the study period.

Subgroup Analyses Separate regression models were fit to subsets of patients in order to compare cost and LOS effects of ADEs according to the clinical manifestations of the ADEs. The most common clinical manifestations were gastrointestinal, primarily nausea and vomiting (n  1071), cutaneous effects of rash, itching, or hives (n  504), respiratory depression (n  112), confusion or agitation (n  97), and urinary retention (n  20). In models of untransformed costs, which included DRG weight and surgery type, a trend was seen toward a lower ADE effect on cost and LOS in patients who

had one category of clinical manifestation compared to patients who had experienced multiple types of clinical manifestations. That is, ADE was associated with $127 higher costs when one clinical manifestation was present compared to $1,185 higher costs when more than one manifestation was present; however, the P value for the variation in effect according to the number of clinical manifestations was not statistically significant (P  0.1). Similarly, cutaneous ADEs were associated with a lower effect on costs than other clinical manifestations (P value for interaction term: 0.04). In models of log-transformed costs and of LOS, the variation of effect according to clinical manifestation was less pronounced. Table 4 shows the linear regression subgroup models by ADE type. The percentage increase in costs, as determined by linear regression models of log transformed cost, range from a low of a 4.2% increase in costs for patients with an opioid related ADE who have a rash, hives, or

Table 3 Regression Models for LOS and Hospital Cost Outcome Length of stay Length of stay Length of stay Cost Cost Cost a Beta

Model Type

Transformation

Adjusted Adverse Drug Event Effecta

95% Confidence Interval

P value

Fixed Fixed Random Fixed Fixed Random

No Log No No Log No

0.47 0.18b 0.85 355 0.15c 786

0.18–0.77 0.15–0.21 0.70–0.99 218–928 0.12–0.18 489–1,084

0.001 0.001 0.001 0.22 0.001 0.0082

coefficient from model that included matched group, surgical procedure category, and DRG-weight. to 20% increase in LOS. c Corresponds to 16% increase in costs. b Corresponds

Vol. 25 No. 3 March 2003

Cost of Opioid Adverse Drug Events

281

Table 4 Linear Regression Subgroup Models for ADE Type ADE Typea Clinical Manifestation Gastrointestinal (nausea/vomiting) Cutaneous (rash/hives/itching) CNS (confusion/lethargy) Respiratory (resp depression) Urinary (urinary retention)

Outcome

Adverse Drug Event Effect

95% Confidence Interval

P value

% Increaseb

LOS (days) log cost LOS (days) log cost LOS (days) log cost LOS (days) log cost LOS (days) log cost

0.47 0.096 0.26 0.041 1.97 0.106 1.35 0.096 4.41 0.135

0.14–0.80 0.042–0.144 0.28–0.80 0.021–0.061 0.39–3.55 0.053–0.158 0.69–2.63 0.049–0.143 0.57–8.24 0.021–0.249

0.005 0.001 0.340 0.001 0.015 0.001 0.039 0.001 0.025 0.020

8.6 10.1 4.6 4.2 23.6 11.2 19.8 10.1 65.2 14.5

Estimated Cost Increase ($)c 510 209 942 626 777

aCategories

of clinical manifestations are not exclusive, namely, a patient with more than one clinical manifestation (e.g., nausea and confusion) would appear in both categories. b % increase for costs were calculated by exponentiating the effect coefficient from the log transformed model. % increase for length of stay was determined by calculating the percent change from the mean in all controls matched to the ADE patient with the specific clinical effect, for example, CNS. c Estimated cost increase was calculated by multiplying the % increase by the median cost all controls matched to the ADE patient with the specific clinical effect, for example, CNS.

itching to a high of 14.5% for patients with urinary retention. Subgroup analyses also were done for the major surgical categories. The matched regression that included DRG weight showed statistically significant cost coefficients (P  0.001) for all surgery types except otolaryngological.

Discussion The problem of ADEs has received considerable attention in recent years, with the increasing recognition paid to patient safety and medical errors. Cost estimates for the impact of ADEs and other types of adverse events have been based on relatively few studies. In addition to confirming results of previous studies, there is a need to perform detailed studies of outcomes associated with ADEs for specific types of drugs and patient populations. In this study, we focused on the impact of ADEs caused by opioids because this class of medications is used with very high frequency in hospitalized patients, particularly for postsurgical analgesia. The results substantiate, with a longer surveillance experience, earlier data from LDS Hospital indicating that opioids are the predominant class of medications associated with ADEs in hospitalized patients. Indeed, opioids accounted for 59% of all ADEs detected during the 10 year period of surveillance. This complication was not limited to any single surgical service but rather was the dominant cause of ADE for each of the major surgi-

cal services. Rates of opioid-related ADEs reported in this study still probably represent an underestimate of their true rate because only moderate or severe adverse events were targeted by the surveillance system. The notable finding of this study was that opioid-associated ADEs were related to both increased length of stay and costs with no effect on mortality. Increased costs correlate with longer length of stay with associated increased health professional time, per diem charges, and direct costs of treating the adverse event including costs related to diagnostic, laboratory, antidote, radiology, and symptom control medications. The magnitude of increase in costs and length of stay observed in this study was less than the average increase in cost and length reported previously, when all ADEs were combined. In two previously published ADE cost analyses, the additional cost per case associated with an ADE was estimated to exceed $3,000 and the additional length of stay exceeded 2 days.1,6 This difference may reflect, in part, differences in methodology, in that this study employed additional levels of confounding control. Probably of greater importance is the distinction that the present study focused only on opioid-related ADEs and had a much larger sample size for this type of ADE than other published reports. Moreover, the vast majority of the opioid-related ADEs in this study were classified as moderate rather than severe, in level of severity. It is worth noting that one of the likely effects of the prospective, computer-based surveillance

282

Oderda et al.

ADE system in place at LDS Hospital is early detection, rapid initiation of corrective action, and mitigation of ADE severity. This study, like other investigations of cost outcomes, raises knotty statistical issues.15 We approached the controversial question of whether to apply logarithmic transformation of costs due to their non-normality by constructing models for both types of endpoints. We also performed analyses in which the matched group was treated as a random effect, a method that relaxes the more stringent assumptions of the fixed effect model. In the non-transformed model, the statistical significance of the ADE effect was reduced, due to the influence of a relatively limited number of outliers. The other cost analyses, which were more robust to skewness, demonstrated effects that were highly statistically significant. We believe that, taken together, the results strongly support the conclusion that opioid-related ADEs have a discernible impact on costs. Although the additional cost per case is relatively limited, extrapolated to the entire hospital, annual costs associated with opioid related ADEs in just surgical patients alone are estimated to exceed $130,000 and needs to be taken into account. Limitations of this study were that it was conducted in one hospital and was restricted to adult patients admitted for surgery. If patients experienced problems with opioids after discharge, that information was not known, and associated costs were not included. Despite the presence of an active, computer and pharmacist-based prospective surveillance system, it is likely that not all ADEs were identified during the 10-year period. Data available for this study did not include outcome measures related to the treatment of pain. We do not know whether the use of opioids eventually led to the successful elimination or minimization of pain. It is possible that side effects could have prevented the patients from receiving adequate pain control. In addition, the very proactive ADE surveillance systems may have resulted in prevention and attenuation of some ADEs; therefore, the results here may be less generalizable and more conservative compared to other institutions.

Conclusions The opioid-associated ADE rate in surgical patients who received opioids at LDS hospital

Vol. 25 No. 3 March 2003

was estimated to be 2.67%, resulting in approximately a one-half day increase in LOS and approximately $840 in additional hospital costs per patient. The attributable increase in LOS and extra costs associated with the occurrence of an opioid-related ADE is significant. Although opioids can provide good pain control, ADEs need to be minimized to improve patients’ overall postoperative experience and limit economic consequences. The findings of this study suggest that we should investigate the active use of opioid dose-sparing strategies such as concurrent nonopioid analgesics, cognitive behavioral techniques and other nonpharmacological interventions such as physical medicine approaches. Such multimodality pain management may help decrease the incidence and severity of opioid-associated adverse events, improve patient comfort and recovery experience, and reduce costs and length of stay.

Acknowledgments Funding for this project was provided by Pharmacia Corporation.

References 1. Committee on Quality of Health Care in America, Institute of Medicine. To Err is Human—Building a Safer Health System. Washington, DC; National Academy Press, 2000. 2. Lazarou J, Pomeranz BH, Corey PN. Incidence of adverse drug reactions in hospitalized patients: a meta-analysis of prospective studies. JAMA 1998; 279(15):1200–1205. 3. Acute Pain Management Guideline Panel. Acute Pain Management: Operative or Medical Procedures and Trauma; Clinical Practice Guideline. AHCPR Publication Number 92-0032, Rockville MD, Agency for Health Care Policy and Research, U.S. Department of Health and Human Services, Public Health Service, 1992. 4. American Pain Society. Principles of analgesic use in the treatment of acute and cancer pain, 4th ed. Glenbrook, IL: American Pain Society, 1999. 5. Bates DW, Spell N, Cullen DJ, et al. The cost of adverse drug events in hospitalized patients. JAMA 1997;277(4):307–311. 6. Classen DC, Pestotnik SL, Evans RS, et al. Adverse drug events in hospitalized patients. Excess length of stay, extra costs, and attributable mortality. JAMA 1997;277:301–306. 7. Thomas EJ, Studdert DM, David M, et al. Costs of medical injuries in Utah and Colorado. Inquiry 1999:36:255–264.

Vol. 25 No. 3 March 2003

Cost of Opioid Adverse Drug Events

8. Pryor TA, Gardner RM, Clayton PD, Warner HR. The HELP system. J Med Sys 1983;7:87–102. 9. Naranjo CA, Busto U, Sellers EM, et al. A method for estimating the probability of adverse drug reactions. Clin Pharmacol Ther 1981;30:239–245.

283

12. Littell RC, Milliken GA, Stroup WW, Wolfinger RD. SAS(R) System for Mixed Models. Cary, NC: SAS Institute Inc., 1996. 13. Stata Corporation, 4905 Lakeway Drive, College Station, TX 77845, www.stata.com.

10. List of Diagnosis Related Groups (DRGs)—1999. http://www.hcfa.gov/stats/medpar/drg99dsc.xls.

14. SAS Institute Inc., SAS Campus Drive Cary, NC 27513-2414, USA. www.sas.com.

11. Standard Cost Manager: User’s Guide, Cleveland, Ohio: Ernst and Whinney, Intermountain Health Care Inc., 1987.

15. Zhou XH, Melfi CA, Hui SL. Methods for comparison of cost data. Ann Intern Med 1997;127(8 Pt 2):752–756.